4 ms·

Are they going to verify that everyone said their sequence correctly? It would be a silly and pointless prank to derail the effort by omitting a number on purp

Are they going to verify that everyone said their sequence correctly? It would be a silly and pointless prank to derail the effort by omitting a number on purpose, but this is the internet... Or maybe it's just the 'coming together' aspect that we're going for anyway, in which case, it doesn't matter :)

aaron695 2y ago
> Are they going to verify that everyone said their sequence correctly? It's trivial-ish to do. First iteration - Google already auto-transcribes, just download the subtitles and compare to the required numbers. Someone else keen can check if that works - https://www.youtube.com/watch?v=qsDAGe7lsII https://www.youtube.com/watch?v=qsDAGe7lsII
LorenDB 2y ago
Maybe they could run it through Whisper on Groq? It should be able to process individual clips in about a second.
- kylecazar 2y ago
  That's what I was thinking... Compare the assigned chunk text with the output from an LLM transcription.
- dudus 2y ago
  Sure and we could also run these LLMs on Arduino s because why not?
- eurleif 2y ago
  Except that: >We're aware it's likely that if you're reading out a long string of digits like this, you're likely to slip up or say a wrong digit. Mistakes are a very human feature, but as long as you correct it and carry on, they're fine - we just need evidence of humans having said the whole prime, not necessarily in one go. You can watch your video back through before submitting to check it's ok. If you want to edit together multiple clips to make your video, that's also fine! So they need the some degree of semantic understanding to recognize corrections, in the various forms they can take. Speech to text alone is insufficient.
  - HappMacDonald 2y ago
    But errors would be rare enough that a single sufficiently OCD soul (or a handful if you wanted error correction of last verification step) could look over the transcripts and diff them against the text of the prime and ensure that every digit in the prime is part of a large sequence of digits read out by the people.
  - withinboredom 2y ago
    You don’t need semantics. Just what is expected. If it deviates and returns to what is expected, it is a success. If it deviates and doesn’t return, it is incorrect. The assertion is that all the digits are said, not that all the digits are said flawlessly.
    - eurleif 2y ago
      Suppose the digit sequence is: "1, 2, 3, 4, 5". Someone says: "9, 2, 3 -- wait, the digit a few places back was a 1, not a 9 -- 4, 5". The pattern /^.*1.*2.*3.*4.*5.*$/ doesn't match; they didn't "deviate and return" in the way you seem to be implying; yet they've expressed the correction in a reasonable way that a human would understand.
      - withinboredom 2y ago
        Depends on how you write the algorithm. As you mentioned, a regex won't work, but "1, 2, 3, 4, 5" is spoken in the clip.
addandsubtract 2y ago
...and accidentally discover the next biggest prime number!

aaron695 2y ago

> Are they going to verify that everyone said their sequence correctly? It's trivial-ish to do. First iteration - Google already auto-transcribes, just download the subtitles and compare to the required numbers. Someone else keen can check if that works - https://www.youtube.com/watch?v=qsDAGe7lsII https://www.youtube.com/watch?v=qsDAGe7lsII

LorenDB 2y ago

Maybe they could run it through Whisper on Groq? It should be able to process individual clips in about a second.

kylecazar 2y ago
That's what I was thinking... Compare the assigned chunk text with the output from an LLM transcription.
dudus 2y ago
Sure and we could also run these LLMs on Arduino s because why not?
eurleif 2y ago
Except that: >We're aware it's likely that if you're reading out a long string of digits like this, you're likely to slip up or say a wrong digit. Mistakes are a very human feature, but as long as you correct it and carry on, they're fine - we just need evidence of humans having said the whole prime, not necessarily in one go. You can watch your video back through before submitting to check it's ok. If you want to edit together multiple clips to make your video, that's also fine! So they need the some degree of semantic understanding to recognize corrections, in the various forms they can take. Speech to text alone is insufficient.
- HappMacDonald 2y ago
  But errors would be rare enough that a single sufficiently OCD soul (or a handful if you wanted error correction of last verification step) could look over the transcripts and diff them against the text of the prime and ensure that every digit in the prime is part of a large sequence of digits read out by the people.
- withinboredom 2y ago
  You don’t need semantics. Just what is expected. If it deviates and returns to what is expected, it is a success. If it deviates and doesn’t return, it is incorrect. The assertion is that all the digits are said, not that all the digits are said flawlessly.
  - eurleif 2y ago
    Suppose the digit sequence is: "1, 2, 3, 4, 5". Someone says: "9, 2, 3 -- wait, the digit a few places back was a 1, not a 9 -- 4, 5". The pattern /^.*1.*2.*3.*4.*5.*$/ doesn't match; they didn't "deviate and return" in the way you seem to be implying; yet they've expressed the correction in a reasonable way that a human would understand.
    - withinboredom 2y ago
      Depends on how you write the algorithm. As you mentioned, a regex won't work, but "1, 2, 3, 4, 5" is spoken in the clip.

addandsubtract 2y ago

...and accidentally discover the next biggest prime number!